Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Update multiple time step deterministic test #2381

Merged
merged 4 commits into from
Apr 15, 2020

Conversation

mcbennet
Copy link
Contributor

Increasing initial timestep.

With an initial timestep of 0.6/Ha, the develop branch prior to #2377 fails for both DMC series. The error is also larger for the first series.

Proposed changes

Changing initial timestep of tests/solids/diamondC_1x1x1_pp/det_qmc_short_vmc_dmc_lg_ts.in.xml from 0.3/Ha to 0.6/Ha

Does this introduce a breaking change?

  • No

What systems has this change been tested on?

Workstation

Checklist

Update the following with a yes where the items apply. If you're unsure about any of them, don't hesitate to ask. This is
simply a reminder of what we are going to look for before merging your code.

  • Yes/No. This PR is up to date with current the current state of 'develop'

Increasing initial timestep.
@qmc-robot
Copy link

Can one of the admins verify this patch?

@ye-luo
Copy link
Contributor

ye-luo commented Apr 15, 2020

I'd like to remove bug tag ASAP so we can track it regularily.
I know that the complex build fails due to fixed phase vs fixed node. We either need to add reference number for complex or skip the test.
Also mixed precision needs to be checked.
@Hyeondeok-Shin please advice.

@prckent prckent changed the title Update Deterministic test Update multiple time step deterministic test Apr 15, 2020
@mcbennet
Copy link
Contributor Author

mcbennet commented Apr 15, 2020

I know that the complex build fails due to fixed phase vs fixed node. We either need to add reference number for complex or skip the test.
Also mixed precision needs to be checked.

I can add these to the present PR

@prckent
Copy link
Contributor

prckent commented Apr 15, 2020

My understanding is that the larger timestep simply makes the test more sensitive, since old versions of the code with the bug were passing even with the shorter timestep. Good.

@mcbennet

Thanks for the update.

I suggest to rename
bug-deterministic-diamondC_1x1x1_pp-vmc-dmc-lg_ts_sdj
as
deterministic-diamondC_1x1x1_pp-vmc-dmc-multi_ts_sdj

or similar. The key thing we are testing is varying timesteps.

Please modify this to only run for builds where it has a chance of passing. Ensure CUDA, complex and mixed precision are all skipped.

Edit: Had not seen your reply 2 min ago.

@ye-luo
Copy link
Contributor

ye-luo commented Apr 15, 2020

Skipping CUDA, complex, mixed precision is an OK solution to me as well. The purpose is not to confuse people with an unintended test failing in these builds.

@Hyeondeok-Shin
Copy link
Contributor

@ye-luo I agree with skipping those tests for now.

@ye-luo
Copy link
Contributor

ye-luo commented Apr 15, 2020

Okay to test

@mcbennet
Copy link
Contributor Author

SoA and AoS appear to diverge for 0.6/Ha case -- for both develop and 3.8.0.
Though I have confirmed that across 3.8.0 and Develop, SoA and AoS agree individually, e.g., for AoS case I see that 3.8.0 and Develop agree and same is true for SoA.

@prckent
Copy link
Contributor

prckent commented Apr 15, 2020

The AoS code failed due to this divergence:

        Start 821: deterministic-diamondC_1x1x1_pp-vmc-dmc-multi_ts_sdj-1-1-2-totenergy
219/279 Test #821: deterministic-diamondC_1x1x1_pp-vmc-dmc-multi_ts_sdj-1-1-2-totenergy .....***Failed    0.08 sec
Tests for series 2
  Testing quantity: LocalEnergy
    reference mean value     : -10.20986762
    reference error bar      :   0.00000100
    computed  mean value     : -10.23546172
    computed  error bar      :   0.00000000
    pass tolerance           :   0.00000300  (  3.00000000 sigma)
    deviation from reference :  -0.02559410  (-25594.09700000 sigma)
    error bar of deviation   :   0.00000100
    significance probability :   1.00000000  (gaussian statistics)
    status of this test      :   fail

Test status: fail

Since the AoS code is not long for the world, you could skip it. Alternatively, add distinct test data, or try playing with the input parameters to avoid the divergence between AoS and SoA (if possible). Your call.

@ye-luo ye-luo merged commit 9867d98 into QMCPACK:develop Apr 15, 2020
@mcbennet mcbennet deleted the add_det_test_big_ts branch April 15, 2020 20:37
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants